A user wonders whether reasoning in latent space, rather than executable chains of thought, might make a difference.
A user asks whether looped models might be less susceptible to distillation attacks because more of their “reasoning” occurs in latent space rather than in executable chains of thought.
2 Sources, first seen ago
—
Not ranked yet
—
Not ranked yet
@moyixI wonder if looped models are less susceptible to distillation attacks since more of the “reasoning” occurs in latent space rather than executable CoT?
@mblondel_mlRT @mblondel_ml: @antirez On their website, they alluded to the fact that they're using continuous diffusion, so presumably there could be…@moyixI wonder if looped models are less susceptible to distillation attacks since more of the “reasoning” occurs in latent space rather than executable CoT?@mblondel_mlRT @mblondel_ml: @antirez On their website, they alluded to the fact that they're using continuous diffusion, so presumably there could be…
